Let me just cut straight to the point: MRP and APS in the end simply don’t work. These approaches are based on what I call “the paradigm of the makeable reality”. They start from a large number of doubtful assumptions about this reality, like demand, inventory, bills-of-material, lot sizes, lead times, reject rates and capacity. And when these assumptions do not come to pass, it leads to nervousness and instability in the system. It’s time for something else. It’s time for an approach that accepts that reality isn’t manufacturable.
The Manufacturable Reality
There is no lack of examples. Planners that after the weekly MRP run are flooded with reschedule in and out messages; sometimes even thousands! During the week they try to expedite all reschedule in lines, but unfortunately they can’t make everything come in on time. Material shortages and line stops are often the consequence. And those reschedule out messages? We really don’t get to these while expediting, so inventories of those parts tend to surge in parallel. But no worries, all of this will be corrected in next week’s MRP run. And of course, we can of course always improve this situation by running MRP on a daily basis…
The causes of this very real scenario: assumptions about reality. Assumptions about demand that turns out to be different, assumptions about reject levels that in reality are always lower or higher, assumptions about available stock that turns out not to be there, and so on. Assumptions that turn out to be wrong and that need (and will be) corrected in the next run. Corrections that often create more variation than actually present in reality. In fact, a very nice illustration of Deming’s famous funnel experiment. In this experiment, Deming shows what the disastrous effects are of managers “tampering” with a system without a proper understanding of variation.
The countermeasure that are typically proposed come down to “do not allow variation”, “better your assumptions” or “re- calculate more often”. So we try to impose so called time fences on our customers, not allowing them to change demand in the coming few weeks or even months (as if reality won’t catch up…). And investing in presumably better forecasting techniques (often going with proportionally expensive, complex and less understood systems) or “advanced” systems with comprehensive, detailed models allowing daily re-calculation, also seems to draw crowds. But in my opinion, we do not address the root cause: thinking reality is makeable.
Lean: a Robust, Reflexive Answer
Lean offers an alternative for the approach described above. The starting point for the short term, operational control of the material flow thereby is reacting to reality instead of anticipating based upon assumptions.
In brief, the concept comes down to “sell one, make one, buy one” and is often realized with visual tools like the well-known kanbans. In this way, Lean tries to achieve continuous flow and a controlled level of work-in-process and thus lead time. It has its theoretical origins in queuing theory, applied to every individual material flow. And from a capacity point of view, it comes down to work load balancing at the most operational level, and based upon the real time state of the work-in-process on the shop floor. Lastly, it also consistent with the ideas in Statistical Process Control (SPC), one of the other basic theories that underly Lean thinking. For instance, in his 1982 book “Out of the Crisis” Deming stated “Kanban implies statistical control of speed of production”.
The way Lean deals with variation and preventing the bullwhip or Forrester-effect is strongly based upon SPC thinking. Briefly stated, in case of common cause variation the policy should be “don’t just do something, stand there”. However, when dealing with special cause variation, the policy should be “don’t just stand there, do something”. From a logistics point of view it implies that we do not interfere (i.e., change our current system of work) when the process is in statistical control (i.e., only shows natural variation). When we would correct (or “tamper” as Deming calls it) based upon this type of variation, it will only lead to even more variation.
So, Let’s Stop Planning Then?
Surely not! Also here, SPC provides us with the basic direction. When our process gets out of statistical control (and it will at some point in time), we actually do need to adjust our system. In fact the control limits of our process are changing, which in our kanban system means that we need to adjust the number of kanbans in our loops and at a larger level may imply adjusting our capacity. Evaluating our process and possibly adjusting our system to the new reality is something we tend to do at the planning level. So surely we still need to plan. However, the real time coordination of the exact timing and quantity of producing and moving specific part numbers is gladly left to the kanban system on the shop floor.
To conclude, some final remarks about the control and the reduction of stocks. As said, kanban aims at realizing a state of statistical control with regards to WIP and lead time. However, the number of kanbans – and therefore the maximum allowable stock – to a large extent is dependent on the level of variation present. Variation that for a part relates to true demand variation, but mostly comes from the upstream limitations and constraints (like change-over time, lot sizes, production intervals, down time, quality issues, and so on). This also clarifies why so many Lean organizations put so much more emphasis on inventory while evaluating their factories and value streams as do more traditional companies. Inventory reduction mostly comes down to tackling our constraints; inventory control to introducing pull flow, managed by the autonomous teams in the value stream. And within the autonomy of these teams it is where we tackle our obstacles on our road towards the ideal of one piece flow – kanban fueling the engine of kaizen!